← CHI 2026 map

CHI '26 · Best paper · full-paper review · confidence high

AI Sensing and Intervention in Higher Education: Student Perceptions of Learning Impacts, Affective Responses, and Ethical Priorities

Bingyi Han , Ying Ma , Simon Coghlan , Dana McKay , George Buchanan , Wally Smith

DOI PDF Program page Page JSON

This best paper matters because it shows the problem is not merely which sensing modality is used, but that disclosed monitoring itself degrades students’ perceived learning experience and emotional comfort. The work also shows students prefer private, system-led help over teacher escalation and rank autonomy and privacy above other ethical values, giving AIEd designers a sharper student-centred brief.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

This paper is a strong empirical intervention in the AIEd conversation because it challenges a common design instinct: if sensing becomes accurate enough and interventions become personalised enough, students will welcome the system. The study instead shows that once students know sensing is happening, their judgments worsen across both affective and learning-related measures, and this holds across gaze-based attention detection and facial emotion detection. That is an important correction to a field that often treats modality choice as the main acceptability question. The more consequential issue here is the social and psychological meaning of being monitored in a learning setting. A second important contribution is the finding that students prefer system-generated hints over teacher-mediated assistance. That preference is not framed as anti-human in a simplistic sense; rather, it reflects concerns about agency, privacy, embarrassment, and being publicly marked as struggling. The qualitative material makes clear that classroom social dynamics matter. Teacher involvement can be experienced not as supportive care but as exposure. This is exactly the kind of contextual insight HCI should surface before educational institutions operationalise sensing-heavy systems. The ethics component is also more than a checklist exercise. By combining ratings with pairwise comparisons, the paper shows that autonomy and privacy are not just present concerns but the most consistently prioritised ones, ahead of fairness, transparency, accuracy, and learning beneficence. That ordering is useful because many AI governance discussions foreground fairness or transparency first, whereas these students emphasise control over their learning process and data. The paper therefore contributes descriptive knowledge with direct design implications: minimise intrusive sensing, preserve student control, prefer private feedback channels, and treat consent and social sensitivity as core system requirements rather than afterthoughts. The main caveat is that the evidence comes from scenario-based reactions by Australian university students, so the results should not be overgeneralised to real deployments, other cultures, or younger learners. Still, as a best-paper contribution, it is persuasive precisely because it identifies a robust pattern that many technically oriented AIEd projects risk missing: monitoring can undermine the very educational benefits it is meant to enable.

What Changed

Canon before

Prior dominant assumption or baseline is that AI sensing and intervention in education is primarily evaluated from a technology-centred perspective focusing on effectiveness, efficiency, and usability, expecting technical performance to translate to positive educational experiences. It assumes different sensing modalities have varying user comfort levels, and that teacher-mediated intervention is preferable due to empathy and personal connection.

Departure from common sense

This paper breaks the assumption that sensing modality matters significantly; students responded negatively to AI monitoring regardless of whether gaze-based or facial-based sensing was used. It also overturns the expectation that teacher-mediated assistance is preferred; students preferred system-generated hints over teacher interventions, citing concerns about autonomy and social embarrassment.

Actual novelty

The study provides novel empirical evidence that disclosed AI sensing itself drives negative student reactions across learning and affective judgments, regardless of whether the system uses gaze or facial sensing. It also contributes a structured ethical-prioritisation result showing autonomy and privacy outrank fairness, transparency, accuracy, and learning beneficence, alongside concrete student-centred design implications for less intrusive intervention systems.

Evidence

Evidence comes from a two-stage mixed-method online study with 132 Australian university students. Stage 1 used video scenarios varying sensing use, sensing modality, and intervention form, with Likert ratings and open-ended responses. Stage 2 used ethical concern ratings and 1980 pairwise comparisons analysed descriptively and with a Bradley-Terry model. The findings consistently show negative reactions to sensing, no modality advantage, preference for system hints over teacher intervention, and prioritisation of autonomy and privacy.

Limits

Method limits

The study relies on video scenarios rather than actual use of AI sensing-intervention systems, so it does not fully simulate the experience of engaging in a real learning activity. It also captures immediate reactions rather than longer-term adaptation.

Deployment limits

The sample is limited to Australian higher education students, so expectations, cultural norms, and ethical concerns may differ across countries, regulatory regimes, age groups, and K-12 settings. Real classroom deployment may also change reactions over time.

Boundary conditions

The claims are best read as applying to hypothetical AI sensing-intervention concepts in higher education, especially where monitoring is disclosed and interventions may expose students socially. Results may vary in other cultural contexts, educational levels, or with actual systems that differ in intrusiveness, consent, and privacy protections.

Position in field

The paper pushes AIEd and HCI away from a purely technology-centred evaluation frame toward a student-centred account of affect, agency, privacy, and classroom social dynamics. Its main contribution is not a new sensing technique but a strong empirical correction to optimistic assumptions about monitoring-based personalisation and a clearer ethical ordering of student concerns.

Abstract

AI technologies that sense student attention and emotions to enable more personalised teaching interventions are increasingly promoted, but raise pressing questions about student learning, wellbeing, and ethics. In particular, students’ perspectives about AI sensing-intervention in learning are often overlooked. We conducted an online mixed-method experiment with Australian university students (N=132), presenting video scenarios varying by whether sensing was used (in-use vs. not-in-use), sensing modality (gaze-based attention detection vs. facial-based emotion detection), and intervention (by digital device vs. teacher). Participants also completed pairwise ranking tasks to prioritise six core ethical concerns. Findings revealed that students valued targeted intervention but responded negatively to AI monitoring, regardless of sensing methods. Students preferred system-generated hints over teacher-initiated assistance, citing learning agency and social embarrassment concerns. Students’ ethical considerations prioritised autonomy and privacy, followed by transparency, accuracy, fairness, and learning beneficence. We advocate designing customisable, social-sensitive, non-intrusive systems that preserve student control, agency, and well-being.